st. louis fed - new seasonal adjustment factors …...update of seasonal adjustment factors...
TRANSCRIPT
New Seasonal Adjustment Factorsfor the Establishment Data Series
Kirk J. Mueller
Twice a year, the Bureau of Labor Statistics (BLS)computes and publishes projected seasonaladjustment factors used to seasonally adjust
establishment-based employment, hours, and earnings data.Tables 1-6 present factors for all published series during the8-month period, September 2000 through April 2001.Revised factors from this update have been used to seasonallyadjust the September final, October second preliminaryestimates, and November first preliminary estimates. As isusual practice, the annual revision of historical seasonallyadjusted data will occur in June 2001, concurrent with therelease of the new benchmarks and the next semi-annualupdate of seasonal adjustment factors (covering March-October 2001).
Seasonal factors in this issue of Employment and Earningswere derived using January 1990 through October 2000 data.The September and October factors replace those publishedin the June 2000 issue of Employment and Earnings.Seasonally adjusted data are not published for those serieswith small or irregular components or both. However, theseseries, shown in tables 1-4, are used in aggregations ofbroader seasonally adjusted levels.
For employment, seasonally adjusted factors are applieddirectly to the 2-digit levels with various seasonally adjustedtotals up through total nonfarm employment derived throughaggregation of the appropriate component series. Seriesbelow the 2-digit level are independently adjusted and notused in aggregations. Seasonally adjusted total private hoursestimates are weighted averages of seasonally adjusted dataat the 2-digit level in manufacturing and division level for
other private industries. Seasonally adjusted total privateearnings estimates are weighted averages of all divisions.
BLS uses X-12 ARIMA (Auto-Regressive IntegratedMoving Average) software, developed by the U.S. CensusBureau, to seasonally adjust the establishment-basedemployment, hours, and earning series.1 All series arecomputed using multiplicative models. The X-12 ARIMAprocess enables BLS to refine its seasonal adjustmentprocedures to control for survey interval variations,sometimes referred to as the 4- versus 5-week effect.
A further refinement—the calendar effect—is made inthe hours and earnings seasonal adjustment that corrects forchanges in the number of weekdays in a month. Thisadjustment is made to all division-level hours' series in theservice-producing sector and the division-level earnings'series for wholesale trade; finance, insurance, and real estate;and services. The series to which the length-of-pay periodadjustment is applied are not subject to the 4- versus 5-weekadjustment, since modeling cannot support the number ofvariables required in the regression equation to make bothadjustments.
Special adjustments for average weekly hours and averageweekly overtime series also are made to account for thepresence or absence of religious holidays in the April surveyreference period and Labor Day in the September referenceperiod. The annual November special adjustment made forpoll workers in the local government (except education)series is also part of this process.
Current seasonal adjustment factors are available on theInternet at: http://stats.bls.gov/ceshome.htm.
Kirk J. Mueller is a supervisory statistician in the Division of MonthlyIndustry Employment Statistics, Bureau of Labor Statistics. Telephone: (202)691-6555; E-mail: [email protected].
1 For a more detailed discussion of the seasonal adjustment procedure,see "BLS Establishment Estimates Revised to Incorporate March 1999Benchmarks" in the June 2000 issue of Employment and Earnings.Additional articles in this series appear in previous June issues.
Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis
December 2000
New Seasonal Adjustment Factorsfor the Establishment Data Series
Christopher D. Manning
Twice a year, the Bureau of Labor Statistics (BLS)computes and publishes projected seasonaladjustment factors used to seasonally adjust
establishment-based employment, hours, and earnings data.Tables 1-6 present factors for all published series during the8-month period, September 2001 through April 2002.Revised factors from this update have been used to seasonallyadjust the September final, October second preliminaryestimates, and November first preliminary estimates. As isusual practice, the annual revision of historical seasonallyadjusted data will occur in June 2002, concurrent with therelease of the new benchmarks and the next semiannualupdate of seasonal adjustment factors (covering March-October 2002).
Seasonal factors in this issue of Employment and Earningswere derived using January 1991 through October 2001 data.The September and October factors replace those publishedin the June 2001 issue of Employment and Earnings.Seasonally adjusted data are not published for those serieswith small or irregular components or both. However, theseseries, shown in tables 1-4, are used in aggregations ofbroader seasonally adjusted levels.
For employment, seasonally adjusted factors are applieddirectly to the 2-digit levels with various seasonally adjustedtotals up through total nonfarm employment derived throughaggregation of the appropriate component series. Seriesbelow the 2-digit level are independently adjusted and notused in aggregations. Seasonally adjusted total private hoursestimates are weighted averages of seasonally adjusted dataat the 2-digit level in manufacturing and division level for
other private industries. Seasonally adjusted total privateearnings estimates are weighted averages of all divisions.
BLS uses X-12 ARIMA (Auto-Regressive IntegratedMoving Average) software, developed by the U.S. CensusBureau, to seasonally adjust the establishment-basedemployment, hours, and earnings series.1 All series arecomputed using multiplicative models. The X-12 ARIMAprocess enables BLS to refine its seasonal adjustmentprocedures to control for survey interval variations,sometimes referred to as the 4- versus 5-week effect.
A further refinement—the calendar effect—is made inthe hours and earnings seasonal adjustment that corrects forchanges in the number of weekdays in a month. Thisadjustment is made to all division-level hours' series in theservice-producing sector and the division-level earnings'series for wholesale trade; finance, insurance, and real estate;and services. The series to which the length-of-pay periodadjustment is applied are not subject to the 4- versus 5-weekadjustment, since modeling cannot support the number ofvariables required in the regression equation to make bothadjustments.
Special adjustments for average weekly hours and averageweekly overtime series also are made to account for thepresence or absence of religious holidays in the April surveyreference period and Labor Day in the September referenceperiod. The annual November special adjustment made forpoll workers in the local government (except education)series is also part of this process.
Current seasonal adjustment factors are available on theInternet at: http://www.bls.gov/ces/cessfin.htm.
Christopher D. Manning is an economist in the Division of CurrentEmployment Statistics, Bureau of Labor Statistics. Telephone: (202)691-6555; e-mail: Manning [email protected].
1 For a more detailed discussion of the seasonal adjustment procedure,see "BLS Establishment Estimates Revised to Incorporate March 2000Benchmarks" in the June 2001 issue of Employment and Earnings.Additional articles in this series appear in previous June issues.
Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis
December 2001
New Seasonal Adjustment Factorsfor the Establishment Data Series
Christopher D. Manning
Twice a year, the Bureau of Labor Statistics (BLS)computes and publishes projected seasonaladjustment factors used to seasonally adjust
establishment-based employment, hours, and earnings dataproduced by the Current Employment Statistics (CES)program. Tables 1-6 present factors for all published seriesduring the 8-month period, September 2002 through April2003. Revised factors from this update have been used toseasonally adjust the September final, October secondpreliminary estimates, and November first preliminaryestimates. As is usual practice, the annual revision ofhistorical seasonally adjusted data will occur in June 2003,concurrent with the release of the new benchmarks. At thattime, the CES program will convert to the use of concurrentseasonal adjustment. Concurrent seasonal adjustment usesall available monthly estimates, including those for thecurrent month, in developing seasonal factors. As notedabove, the CES program currently projects and publishesseasonal factors twice a year. With the introduction ofconcurrent seasonal adjustment, BLS will no longer publishprojected seasonal factors for CES national series. Inaddition, June 2003 will mark the conversion of nationalnonfarm payroll series from the 1987 Standard IndustrialClassification System to the 2002 North American IndustryClassification System.
Seasonal factors in this issue of Employment and Earningswere derived using January 1992 through October 2002 data.The September and October factors replace those publishedin the June 2002 issue of Employment and Earnings.Seasonally adjusted data are not published for those serieswith small or irregular components or both. However, theseseries, shown in tables 1-4, are used in aggregations of broaderseasonally adjusted levels.
For employment, seasonally adjusted factors are applieddirectly to the 2-digit levels with various seasonally adjustedtotals up through total nonfarm employment derived through
Christopher D. Manning is a senior economist in the Division ofCurrent Employment Statistics, Bureau of Labor Statistics. Telephone:(202) 691-6555; e-mail: [email protected].
aggregation of the appropriate component series. Series belowthe 2-digit level are independently adjusted and notused in aggregations. Seasonally adjusted total private hoursestimates are weighted averages of seasonally adjusted dataat the 2-digit level in manufacturing and division levelfor other private industries. Seasonally adjusted totalprivate earnings estimates are weighted averages of alldivisions.
BLS uses X-12 ARIMA (Auto-Regressive IntegratedMoving Average) software, developed by the U.S. CensusBureau, to seasonally adjust the establishment-basedemployment, hours, and earnings series.1 All series arecomputed using multiplicative models. The X-12 ARIMAprocess enables BLS to refine its seasonal adjustmentprocedures to control for survey interval variations,sometimes referred to as the 4- versus 5-week effect.
A further refinement—the calendar effect—is made inthe hours and earnings seasonal adjustment that corrects forchanges in the number of weekdays in a month. Thisadjustment is made to all division-level hours' series in theservice-producing sector and the division-level earnings'series for wholesale trade; finance, insurance, and real estate;and services. The series to which the length-of-pay periodadjustment is applied are not subject to the 4- versus 5-weekadjustment, since modeling cannot support the number ofvariables required in the regression equation to make bothadjustments.
Special adjustments for average weekly hours and averageweekly overtime series also are made to account for thepresence or absence of religious holidays in the April surveyreference period and Labor Day in the September referenceperiod. The annual November special adjustment made forpoll workers in the local government (except education)series is also part of this process.
Current seasonal adjustment factors are available on theInternet at: http://www.bls.gov/ces/cessfin.htm.
1 For a more detailed discussion of the seasonal adjustment procedure,see "BLS Establishment Estimates Revised to Incorporate March 2001Benchmarks" in the June 2002 issue of Employment and Earnings.Additional articles in this series appear in previous June issues.
Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis
December 2002